TradingAgents is an open-source framework that structures several large language model agents to replicate the organization of a traditional trading desk. Rather than relying on a single model to make investment calls, the project splits tasks among specialized agents: fundamental analysts, technical analysts, sentiment analysts, researchers who debate positions, traders who execute decisions, and a risk management module that validates or adjusts proposals before execution.
This multi-agent architecture is directly modeled on how real trading floors operate, where different areas of expertise are weighed against each other before a final call is made. Each LLM agent is assigned a role and specific inputs — technical indicators, financial news, fundamental reports — and produces an analysis that feeds into a structured discussion among agents, intended to reduce the biases of isolated reasoning.
The GitHub repository, maintained by the research group TauricResearch, provides the source code along with backtesting tools to evaluate the system's performance on historical data. The project fits into a broader trend of agentic frameworks applied to finance, where several academic and industry teams are exploring whether LLMs can collaborate effectively on complex, sequential decision-making tasks.
For now, this kind of initiative remains largely experimental: backtesting results do not guarantee performance under live market conditions, and using LLMs for algorithmic trading raises questions around reliability, latency, and risk management that go beyond a research prototype. TradingAgents is primarily aimed at developers and researchers interested in experimenting with multi-agent architectures, rather than at users looking for a ready-to-use trading solution.